kotml is aiming to be a machine learning library written purely in Kotlin. I am aiming to provide an interface similar to sklearn's classifiers. The motivation behind this project was to learn Kotlin and brush up on my ML knowledge.
To get started, include jcenter() in your build gradle repositories.
Then add the following to your dependencies:
compile 'com.tommypacker:kotml:0.1-ALPHA'
You can also use kotml in your maven projects by including the following in your pom.xml:
<dependency>
<groupId>com.tommypacker</groupId>
<artifactId>kotml</artifactId>
<version>0.1-ALPHA</version>
<type>pom</type>
</dependency>
- Naive Bayes classifier
- Gaussian
- Multinomial
- Bernoulli
- K Nearest Neighbors
- Non-Weighted
- Weighted
- Linear Regression with gradient descent (currently in alpha)
- DataContainer class to easily manipulate data given in CSV form. Splits data automatically into testing and training data with a user defined split ratio. Also allows ignoring the first column of datasets in case of an ID column.
- Planned support for many other popular ML models (SVM, Regression, etc).
kotml currently supports three Naive Bayes classifiers(BernoulliNB(), GaussianNB(),
MultinomialNB()), K Nearest Neighbors (KNN()), and LinearRegression (LinearRegression()).
// Load Data
val spamData = Datasets.loadSpam(false, 0.8)
// Create Classifier
val BNB = BernoulliNB()
// Fit classifier to training data and training labels
BNB.fit(spamData.trainingData, spamData.trainingLabels)
// Test model and print accuracy
val accuracy = BNB.test(spamData.testData, spamData.testLabels)
println(accuracy)// Load Regression Data
val regressionData = Datasets.loadSampleRegression()
// Create Model
val sl = LinearRegression()
// Train model for a given learning rate and number of epochs
sl.train(regressionData.trainingData, regressionData.trainingResponses, 0.0005, 10000)
// Make Predictions
val predictedValues = sl.predictValues(regressionData.trainingData)More examples can be found in examples.kt
APIs used in this project:
Guides to help me get started: